ARHIS Ambient Achieves 96.7% Detection Accuracy Rate for Black Ice Detection with Multimodal AI

Black ice, a major cause of winter accidents, is difficult to identify with the naked eye, meaning existing camera-based technology alone has limitations. ARHIS Ambient was developed to solve this problem, and it is a multimodal ice detection solution that simultaneously analyzes driving sound (Audio), video (Image), and environmental information (temperature and humidity).

Environment-Adaptive Weight Adjustment

The core of the solution is the EAL (Environmental Attention Layer). This layer receives environmental data as input and dynamically adjusts the reliability of audio and image features. For example, at night or in situations where visibility is poor, it reduces the weight of video information and increases the weight of audio features to predict icing risk.

Ensuring Stability through Late Fusion

Each modality is independently encoded and then combined via Late Fusion. Audio data, which analyzes the frequency characteristics of tire and road surface friction sounds, compensates for video noise caused by reduced lighting or fog. Conversely, in stopped or congested sections, video and environmental data are weighted more heavily to maintain accuracy.

Performance Verification and Patents

This technology has built a patent portfolio of over 60 patents and has secured a registered US patent (US 12,442,796). In a public institution performance evaluation in October 2025, a comparison of inference results with actual road surface conditions over 25 days recorded a detection accuracy rate (TDR) of 96.7%, officially recognizing its detection performance.